Abstract
Purpose :
Visual acuity (VA) remains the primary functional endpoint for quantifying treatment effectiveness. To enrich VA information collected in retina trials, we apply a novel quantitative VA (qVA) framework to estimate VA thresholds from different testing paradigms (1). We propose that novel Bayesian analytics can reduce the uncertainty of VA estimates in patients with retinal disease.
Methods :
Prospective, observational study performed at Mass Eye and Ear. We recruited patients with vision ranging from 20/15 to 20/100 during regular retina clinic visits. ETDRS testing, in the right and then left eye, was followed by qVA testing, which consisted of 15 sequential rows of 3 optotypes. Patients were retested in the opposite order after at least 30 minutes. We generate a qVA profile using a Bayesian model of VA that defines the probabilities for correctly reporting the different numbers of optotypes presented in ETDRS/qVA tasks. To quantify the evolution of a qVA profile, we calculate the half-width of 68.2% credible intervals (HWCI) for VA threshold estimates (2), as a function of 1-14 rows completed during ETDRS/qVA testing.
Results :
53 eyes of 31 patients, with mean age of 65.1 + 13.3 and 21 females (68%) were included in the study. Repeat testing generated a total of 106 ETDRS/qVA tests. Figure 1 presents the Bayesian priors for VA threshold, and final posteriors for two patients. Figure 2a demonstrates the rapid reduction in average HWCI during VA testing. Figure 2b presents HWCIs for an aggregate analysis concatenating two ETDRS/qVA runs for each patient (N=53). Paired t-tests revealed statistically significant HCWI reductions for qVA: 22% for single and 20% for double runs (p<10-6).
Conclusions :
We demonstrate that the novel application of a qVA analysis reduces uncertainty in VA estimates from ETDRS/qVA testing. The qVA advantage most likely emerges from its intelligent sampling and fine-grain resolution, relative to coarse, static sampling of ETDRS. These results support the feasibility for qVA testing and analysis to improve the signal-noise features of VA data in clinical trials.
References:
Lesmes & Dorr (2019) https://doi.org/10.475/123_4
Zhao et al (2021) https://doi.org/10.1167/tvst.10.1.1
This is a 2021 ARVO Annual Meeting abstract.